Mistral Large 4: score and benchmark evidence
Historical release · · Mistral AI
81.0 points · Rank 32 among 34 ranked models · Stability range 72.2–90 points.
Rank follows the point estimate; it does not establish statistically significant superiority. The scale is anchored at mean 100, SD 15 in the frozen calibration cohort, not human IQ or a percentage.
One score summarizes demonstrated capability across published reasoning settings and seven equally weighted domains. Related benchmark variants share a family budget, and influence from one evaluator is capped.
Release tli-2026-v1.0-2026-10-19-refresh-31. Methodology and limitations for this release.
This page preserves historical evidence. See the current model record.
Capability domains
Domain scores use calibration-cohort standard-deviation units. Unmeasured domains are shown as unavailable, not zero.
| Domain | Score (SD units) | Observed benchmarks |
|---|---|---|
| Knowledge & reasoning | -0.85 | 3 |
| Coding & software engineering | -0.42 | 3 |
| Agentic tool & computer use | -0.36 | 1 |
| Professional & real-world work | -0.55 | 6 |
| Multimodal & vision | -0.57 | 1 |
| Cybersecurity | Unavailable | 0 |
| Preference & communication | Unavailable | 0 |
Published benchmark evidence
14 results contribute to this score, from 22 collected results and 14 contributing benchmark families. Counts are not independent sample sizes.
Results, source citations, review decisions and compatible reasoning modes are stored in a versioned relational database. Each publication uses a sealed, checksum-verified export; CSV files are optional exports, not editable scoring inputs. Previous revisions remain available for audit. Collected means a numeric result exists in the database. Used means its source, exact model checkpoint and benchmark protocol have been reviewed and its benchmark has a frozen calibration supported by multiple providers. For each benchmark we select the highest published aggregate across comparable reasoning modes, not necessarily maximum effort, and count it once. More tested settings provide more chances to record a high score; this selection advantage is not corrected by the conditional interval. Developer-reported and sole/unspecified settings remain eligible and disclosed. Unresolved does not mean false or useless: conflicting metrics, unverified citations and unmatched versions stay collected until reconciled.
The score stability range is a 95% bootstrap score interval conditional on frozen calibration and comparable evidence availability. It resamples benchmark families and evaluators and adds a frozen missing-evidence correction. It excludes selection across reasoning settings, selective publication, calibration changes and new domains. It is not a 95% guarantee about true capability or where a future score will land. Sparse or conflicting evidence can widen the range.
Reporting sensitivity: ±17.6 points. Sensitivity to selectively published results. This stress range is not a confidence interval or a probability statement.
AA-Briefcase v1.1 (Elo): 1391.97 native-points
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-17; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-17
Daily evaluator refresh; replaces 1393.22 recorded 2026-10-14.
GDPval-AA v2.1 (Elo): 1422.17 native-points
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-14; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-14
Daily evaluator refresh; replaces 1422.87 recorded 2026-10-13.
AutomationBench-AA: 59.9013624681%
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
Terminal-Bench 4.0 — AA harness: 26.7676767677%
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
SciCode: 54.1666666667%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
HLE text-only — AA harness: 35.0324374421%
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
GDP.pdf — AA harness: 18.6%
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
CritPt: 10.5714285714%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
AA-Omniscience: -5.3 native-points
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
AA-LCR v1.1: 81.3333333333%
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Exact model and effort in the evaluator chart payload. Preview release; scores may change at general availability.
MMMU-Pro: 76.4161849711%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-06
Independent MMMU-Pro aggregate for this effort. Preview release; scores may change at general availability.
AA Intelligence Index v4.3.2 (score): 38.3770408917 native-points
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (none); Artificial Analysis independent evaluation, 2026-10-06; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-06
Reference composite only, never scored alongside its component evaluations. Preview release; scores may change at general availability.
BioMysteryBench v1 — Vals: 67.04%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI BioMysteryBench harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 67.04 ± 2.67, rank 20/27. Vals notes some benchmarks may use a different provider or parameters.
Code Migration †: 30.56%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI Code Migration harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 30.56 ± 4.22, rank 41/77. Vals notes some benchmarks may use a different provider or parameters.
Excel Modeling Benchmark — Vals: 56.21%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI EMB harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 56.21 ± 3.22, rank 46/74. Vals notes some benchmarks may use a different provider or parameters.
Finance Agent v2 — Vals: 54.68%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI Finance Agent (v2) harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 54.68 ± 0.58, rank 24/78. Vals notes some benchmarks may use a different provider or parameters.
Harvey LAB (Vals): 15.83%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI Harvey's Legal Agent Benchmark harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 15.83 ± 2.96, rank 9/79. Vals notes some benchmarks may use a different provider or parameters.
Legal Research Bench — Vals: 31.73%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI Legal Research Bench harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 31.73 ± 3.23, rank 42/78. Vals notes some benchmarks may use a different provider or parameters.
MedCode — Vals: 40.69%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI MedCode harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 40.69 ± 2.17, rank 61/107. Vals notes some benchmarks may use a different provider or parameters.
MedScribe — Vals: 80.43%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI MedScribe harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 80.43 ± 2.0, rank 57/109. Vals notes some benchmarks may use a different provider or parameters.
ProofBench v1.1 — Vals: 10%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI ProofBench v1.1 harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 10.0 ± 3.02, rank 45/50. Vals notes some benchmarks may use a different provider or parameters.
Vibe Code Bench v1.1 — Vals: 78.4%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Mistral Large 4 Preview (high); Vals AI Vibe Code Bench v1.1 harness; provider Mistral AI; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 78.4 ± 3.55, rank 28/112. Vals notes some benchmarks may use a different provider or parameters.
Cite this model record
The Latent. “Mistral Large 4: score and benchmark evidence.” 2026-10-19. Release tli-2026-v1.0-2026-10-19-refresh-31.
Download release data and definitions (JSON) · Data usage terms